[mlir][sparse] revert optimization for dense->csc conversion.

Eliminates the sort seems make the whole conversion slower (probably because loop rotation leads to bad locality).

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D144517